I am running a simple two-group t-test using modulated, optimized gray
matter data. I have been trying multiple smoothings to see the effects on
small vs large structures. So far I've tried no smoothing, 4, 8, and 12
(the modulated data are 1*1*1mm voxels). However, I've noticed a weird
effect and want to see if i made an inputting blunder somewhere along the
way. In the final results viewer, I get the following numbers:
No Smoothing:
Smooth FWHM = 10 10.5 8
Search volume = 92174 mm^3 / 41.8 resels
voxel = 1 1 1 (1 resel = 843.22 voxels)
***4 Smoothing:
Smooth FWHM = 7.3 8.7 6.8
Search volume = 817746 mm^3 / 1508 resels
voxel = 1 1 1 (1 resel = 429.53 voxels)
8 Smoothing:
Smooth FWHM = 12.6 14.6 12.9
Search volume = 1341375 mm^3 / 512 resels
voxel = 1 1 1 (1 resel = 2383.94 voxels)
12 Smoothing:
Smooth FWHM = 20.1 22.6 20.4
Search volume = 1645251 mm^3 / 167.8 resels
voxel = 1 1 1 (1 resel = 9213.6 voxels)
My issue is that the 4 kernel breaks the general linear trend of the resel
size. It makes sense to me that as you smooth the data, the size of a
resel gets bigger so the number of resels in the image decreases. But why
is the 4 kernel so different from the others? I've tried re-smoothing to
4, and get identical results. Alternately, is something wrong with the
unsmoothed data, or perhaps does the relative smoothness actually indicate
that no smoothing is needed?
Thanks,
Chris
|